Ask ITSD to build a test and production server

[bourque]

Vera Gibbs suggests that we request a test server and environment sooner rather than later, as it helps ITSD plan their work better. The test environment should mimic the production environment, so we should do some thinking on what we would like our production environment to be first.

[SaOgaz]

@bourque, do you mean anything outside of the conda env stuff we might need?

[bourque]

@SaraOgaz I'm referring to another VM machine with the specs that we would want in a production server. As you elude to, we will also need to consider having a jwqlprod conda environment, akin to our jwqldev environment.

[bourque]

Here is what I'm thinking:

Memory

With some JWST files being on the order of 10 GB in size, and with the possibility of several automated scripts reading in files simultaneously, we should have at least ~100GB of RAM. ITSD gave us 128GB for our dev machine so we can get away with having at least that much. Most of the available INS servers have 378GB of RAM, so I think we should ask for that.

Cores

Conceivably we could have several automated scripts using multiprocessing running at once. The more cores the better. Some of the INS servers have 40 CPUs, others have 32, one has 24. I think we should ask for 40.

Accounts

We will need to have our admin account be a member of these servers.

UNIX groups

We will need a prod and test UNIX group, similar to our existing dev UNIX group

[bourque]

A request of Test and Production servers has been sent out!

[bourque]

I've talked with Thomas Walker on this. He indicated that our requests are asking for too many resources from a single VM. He advised me that it is much easier on ITSD's end if we ask for many smaller VMs instead of one large one. We will have to design a better system.

[bourque]

Here is what I'm thinking:

Machines

We can't have one large machine, so we need to split up our needs into several VMs. One school of thought is to have 6 identical machines, one for each instrument plus one for cross-instrument or jwql specific work. Another school of thought is to have a few larger VMs and a few smaller VMs and to do some bookkeeping on our end to figure out which processes should go to which VM.

Memory

There may be some automated scripts that will have to operate on several large JWST files at once, so we should have at least one VM with 64 GB RAM.

Cores

We should ask for at least one VM with 20 cores for larger processes that use multiprocessing

Accounts

Our admin account will need to exist on every VM.

UNIX groups

We will need a prod and test UNIX group, similar to our existing dev UNIX group

Conclusion

Here is what I think we should request:

prod1: 64 GB RAM, 20 Cores
prod2: 64 GB RAM, 20 Cores
prod3: 32 GB RAM, 16 Cores
prod4: 32 GB RAM, 16 Cores
prod5: 24 GB RAM, 8 Cores
prod6: 24 GB RAM, 8 Cores

Totals:
6 VMs, 240 GB RAM, 88 Cores

[bourque]

I have sent in a request for the test server. I was advised by Thomas Walker to wait until this machine is configured to our liking before submitting requests for the production servers.

[bourque]

The test server has been built, and myself and other members are able to log in successfully. I will send in the request for the production servers.

[bourque]

The production servers have been built and seem to work as expected.